DrugCentral sometimes combines chemicals that shouldn't be combined

[gaurav]

For example, an approval for HERCESSI (trastuzumab-strf, UNII:P188ANX8CK) clearly states:

Do not substitute HERCESSI TM(trastuzumab-strf) for or with ado-trastuzumab emtansine or fam-trastuzumab deruxtecan.

However, they are joined by CHEMBL.COMPOUND:CHEMBL1201585 "TRASTUZUMAB", which DrugBank cross-references to both DrugCentral:4979 "trastuzumab" (which is cross-referenced to UNII:P188ANX8CK "TRASTUZUMAB") and DrugCentral:4990 "trastuzumab emtansine" (which is cross-referenced to UNII:SE2KH7T06F "TRASTUZUMAB EMTANSINE").

[gaurav@ht1 DrugCentral]$ grep -r "\b4979\b" xrefs | grep UNII
1751081	P188ANX8CK	UNII	4979	None
[gaurav@ht1 DrugCentral]$ grep -r "\b4979\b" xrefs | grep ChEMBL
1802226	CHEMBL1201585	ChEMBL_ID	4979	None
[gaurav@ht1 DrugCentral]$ grep -r "\bCHEMBL1201585\b" xrefs | grep ChEMBL
1802226	CHEMBL1201585	ChEMBL_ID	4979	None
1807328	CHEMBL1201585	ChEMBL_ID	4990	None
[gaurav@ht1 DrugCentral]$ grep -r "\b4990\b" xrefs | grep UNII
1751071	SE2KH7T06F	UNII	4990	None

We use the DrugCentral public PostgreSQL instance to get xrefs from it, so hopefully we might be able to find some other information from the database to filter some of these out. If not, we might have to remove CHEMBL.COMPOUND mappings from DrugCentral, or reject any crossreferences from DrugCentral that combine two UNII identifiers.

Reported by @gglusman